Getting down to grid computing

Despite its shining promise for establishing elegant, integrated, and cost-effective distributed environments, grid computing has yet to win over skeptical enterprise IT executives and reach mainstream adoption.

Already established as a building block that supports high-performance computing environments in universities and scientific research institutes, grids are now the focus of industry giants. Hewlett-Packard, IBM, Oracle, and Sun Microsystems are betting they can transplant the technology into the hearts of their largest enterprise customers. But concerns that the technology is not mature, its financial track record is sketchy, and even its definition remains unclear keep many corporate users from taking the first steps toward deployment.

“The technical people are somewhat confused [by grids] because vendors are all using a different language despite the fact they are just talking about the same concept — the next wave of distributed computing,” says Dan Kusnetzky, vice president of system software at IDC.

Larry Sikon, CIO at investment bank Thomas Weisel Partners, personifies the cautious view of grids held by many corporate users. “I’m content with the applications I have in place at the moment,” Sikon says. “But [grids] are a neat concept as far as being able to tap spare CPUs, and when I have an application that might apply, such as number crunching, I would consider it.”

In much the same way as an electrical grid does for electricity users, grid computing promises to more efficiently link and provision resources across enterprise platforms. But it may take several years before its value is appreciated, says Mary Johnston Turner, vice president and practice director at research firm Summit Strategies, in an October 2003 report on enterprise grids. “Our research indicates that customers are curious about grids and want to lean more,” Johnson says. “But the jury is still out when it comes to making broad architectural commitments to it. Grid purchasing decisions will be driven by CIOs and architects looking to save money, improve their services level, and increase IT flexibility.”

The top suppliers of grid technologies, predictably, believe grids are the best way to pursue a short-term strategy such as low-level integration of departmental servers, or as a step toward creating a full-blown utility computing environment using platforms such as HP’s Adaptive Enterprise, IBM’s On Demand, Oracle’s flagship product Oracle 10G, and Sun’s N1.

IBM’s companywide On Demand program is particularly grid-centric, with the company using its WebSphere and Tivoli products to conduct policy-based management. Sun sees a grid as fundamentally offering services that provide a policy-based management via its N1 platform. HP will use its OpenView Platform as the basis of its grid strategy. The company will integrate its Talking Blocks Web services management into that platform. Oracle will build its grid hopes around its Oracle 10G database to offer infrastructure provisioning and workload management.

First Steps

One hurdle must be overcome before corporate users will move to grid technology: a remedy for the lack of fully exploitive systems management and security products that can be smoothly melded with existing enterprise infrastructure. Executives with leading grid suppliers know it is critical to address this need before meaningful grid projects can go forward. Nick van der Zweep, HP’s director of virtualization and utility computing, says his company is trying to address questions about manageability. “Right now grids are just APIs, and the management systems available can’t reach in to understand what is going on inside of them,” he says. “Web services management, for instance, allows [HP’s] OpenView to reach in there and understand what is going on the inside of Web services on a grid.”

However, a flood of low cost but fairly robust Intel-based hardware and Linux-based software platforms has been arriving during the past year, and can serve as good entry points for IT shops looking to do their first grid implementations, analysts say.

One of the early commercial markets top-tier grid suppliers are pursuing is financial services. Many bread and butter applications used by that industry involve heavy number crunching and are typically used or accessed by thousands of their business partners and clients.

Hewitt Associates, a global HR company, is deploying an IBM WebSphere-based grid along with grid management software from DataSynapse to run a pension calculations plan, says Dan Kaberon, director of computer resource management. “We’ve achieved great operational stability, which is impressive for such a young enterprise technology as applied to transaction processing,” Kaberon says.

Last month the investment and banking institution Charles Schwab, working in concert with IBM’s High Volume Website Team, brought its grid computing project live, company officials say. The project was designed to improve the performance of its computationally intensive advice applications and to manage business applications running on a variety of different servers. An application included in Schwab’s grid simplifies and automates complex scheduling of workflows.

Targeting Efficiency

Early adopters are finding grids useful for exploiting the idle resources of existing servers, thereby saving money on new purchases and driving greater productivity.

“The economic aspects of grid computing are just beginning to be studied, like the way you might price some compute capacity available to you, or the impact of network speed and latency and the value of that compute capacity,” says Shahin Kahn, vice president of the high performance and technical computing business unit at Sun.

IBM also believes tapping idle servers alone is justification for establishing a grid. “Most studies tell you that Windows servers are typically utilized between 5 percent and 10 percent, with Unix and Linux servers averaging between 15 percent and 20 percent,” says Dan Powers, vice president of IBM’s grid computing strategy. “But look at the mainframe. It has a much higher utilization rate because it has a superior operating system that can virtualize every application below it. This needs to happen on these lower- end platforms.”

IBM continues to expand the technical aspects of its grid strategy through a number of its business partners. One such partner, Avaki, has come up with the “data grid,” in which a single service can fetch data from multiple sources and deliver it customized to a particular developer or user.

Powers believes there are three areas in which corporate users can first dip their toes in the grid computing waters. The first is in using grids to schedule jobs within the network more efficiently. Another involves orchestrated provisioning of important server-based tasks. The final is information virtualization, through which grids can make all the servers flung across an enterprise appear as if it is one big server.

Integration Models

A more general area that has piqued the interest of corporate users is integration. By weaving together an assortment of grid technologies with Web services, some are finding it to be an efficient and cost-effective way of implementing a grid architecture. It is also way of addressing the thorny problem of integrating geographically disparate data stores that would better enable an on-demand e-business.

“If the grid vendors do their jobs properly, users should be able to take applications running in silos and integrate them on grids,” says Dana Gardner, a senior analyst at The Yankee Group. “You should not have to do a major reconfiguration or recompile of your apps.”

The first step is to simplify existing infrastructure by consolidating as many datacenters as possible that will run all IT resources. That done, users should look to consolidate their midrange servers down to a common pool of application servers. Most also advise the consolidation of commonly used applications, such as order processing.

Designing Grid Platforms

With these actions taken, users must then make some hard decisions about what platforms to standardize on and which ones need to go. In so doing, they can significantly simplify implementation, support, and management issues as they layer in grid technologies. Automation becomes attractive as more and more servers and software are added to the original grid. Automation mainly centers around systems and data management.

“Once you get past a cluster of four to eight servers, the challenge becomes to manage it,” says Robert Shimp, Oracle’s vice president of technology marketing. “You need a management infrastructure for the entire datacenter that lets you treat all these servers identically.”

This approach, Shimp contends, affords users the option of starting small, and adding only those management and security pieces they need as they grow. They can then add software for functions such as load balancing, for instance, so that resources can be equally shared across a grid.

Sun’s Kahn believes that an early grid implementation must have the capability of automatically discovering the resources available across that grid, must smoothly support software portability and mobility, must have the innate capability to match workloads to available resources on the fly, and must automatically provision workloads and other resources.

Grids, in Kahn’s scenario, offer significant promise. No doubt IT users will be watching to see if it is fulfilled in the coming years.

“With this in place,” says Kahn, “a grid becomes fertile ground for collaboration and development and gives you a very healthy road map to take you several years out.”